[llvm-commits] [llvm] r161074 - in /llvm/trunk/lib/CodeGen: MachineTraceMetrics.cpp MachineTraceMetrics.h

[Jakob Stoklund Olesen]

On Jul 31, 2012, at 5:19 PM, Andrew Trick <atrick at apple.com> wrote:

> 
> On Jul 31, 2012, at 5:05 PM, Jakob Stoklund Olesen <stoklund at 2pi.dk> wrote:
>>> The SSA assumption should be explicit. Can we change getVRegDef to return MachineOperand?
>> 
>> No, it's used all over the place, and MachineOperand doesn't know its own number anyway.

[MachineOperand doesn't currently have a getOperandNo() method, probably because MO.getParent() can be null.]

>> It this something you need over in MISched? Otherwise I won't add it to MRI.
> 
> No, I'm fine with assertions, or comments to the effect that an assertion elsewhere covers it.

I think I'll add a DataDep constructor with all the assertions.

> I forgot to mention that you're not recursing on phi nodes for instruction depth, so you might end up asking for the latency of a phi node.

Nope, PHIs are transient:

        if (!Dep.DefMI->isTransient())
          DepCycle += MTM.TII->computeOperandLatency(MTM.ItinData,
                                                     Dep.DefMI, Dep.DefOp,
                                                     UseMI, Dep.UseOp,
                                                     /* FindMin = */ false);
 
But I do compute the depth of PHIs, so they do appear as UseMI above. Same thing for copies.

I guess you could resolve both PHIs and copies completely so computeOperandLatency() would get an exact UseMI.

/jakob